LeafDeSNet: A MultiClass plant leaf diseases classification model with entropy-controlled GLEO for feature selection

Plant diseases pose a significant risk to global nutrition and can have a severe impact on small-scale farmers who rely on their crops for survival. Early and accurate detection of plant diseases is essential, yet traditional identification methods are often time-intensive and prone to human error. The development of Computer-Aided Diagnostic (CAD) systems facilitates the early detection of plant diseases for both farmers and experts. These sets of intelligent systems utilize machine learning and computer vision-based techniques to identify and categorize leaf diseases accurately. Such automated approaches not only save time and reduce labor costs but also minimize crop losses by optimizing the yield. This article presents a comprehensive framework for leaf disease classification of three main crops, beginning with image acquisition, proceeding to feature extraction and selection, and concluding with classification. The existence of redundant and irrelevant feature information leads to the problem of “ curse of dimensionality ”. To address this challenge, a bio-inspired optimization approach, known as the Entropy-Controlled Generalized Learning Equilibrium Optimizer (E-CGLEO), is proposed. Unlike the standard GLEO, we used the entropy-based technique to select more diverse features. The conventional GLEO had various constraints that are effectively addressed by our proposed approach: (1) minimal diversity, (2) selection of redundant feature information, and (3) selection based on structural contribution, leading to overfitting. The proposed feature selection framework successfully addresses the identified problems by modifying the objective function and equilibrium condition, while also updating velocity and position, thereby enhancing performance in terms of accuracy, precision, sensitivity, and F1-score.